Source-linked AI summary
Decision Making and Ride Comfort Evaluation Based on Lane-Change Prediction for Autonomous Vehicles
Mohamed Manzour, Aditya Kumar, Rubén Izquierdo, Miguel Ángel Sotelo
TL;DR
Unexpected lane changes can leave little room for braking or avoidance, making the choice of response relevant to both safety and comfort. The study evaluates prediction-informed driving strategies and finds that both predictive responses avoided the collision seen with reactive driving, with lower peak deceleration.
Problem
Unexpected merges into occupied lanes can leave little room to brake or avoid, and the safety and comfort implications depend on the chosen response.
Source: §1 · p. 1Method
An existing knowledge-graph-embedding and Bayesian prediction framework informs CARLA comparisons of reactive, predictive longitudinal, and predictive lateral control, with predicted destination lanes included in lane-risk decisions.
Source: §1 · p. 2, §3.2 · p. 2Results
31.5% and 91.0% reductions in peak longitudinal deceleration with anticipatory braking and lateral avoidance, respectively, accompanied collision avoidance in both predictive configurations.
Source: §5 · p. 6Takeaways & Limitations
Ride-comfort comparisons should account for both motion directions: lateral avoidance lowers combined measures despite increasing lateral acceleration and jerk.
Source: §4.2 · pp. 3–4The comfort result is only an indicative horizontal estimate: it uses simulated vehicle accelerations as passenger proxies and does not fully represent vertical, rotational, or high-frequency vibrations.
Source: §3.4 · p. 3
Abstract
from arXiv · showhide
Predicting surrounding vehicles' lane-change intentions enables autonomous vehicles to anticipate potential hazards and respond proactively rather than relying solely on reactive maneuvers. This study investigates the benefits of integrating lane-change prediction, based on knowledge graph embeddings and Bayesian inference, into autonomous vehicle decision-making. Three driving strategies are evaluated in risky merging scenarios simulated in CARLA: reactive driving, prediction-based longitudinal action, and prediction-based lateral action. The proposed prediction model anticipates the surrounding vehicle's lane change approximately 3-4 seconds before its occurrence, allowing the ego vehicle to adjust its behavior accordingly. In the representative simulation records, reactive driving results in a collision accompanied by aggressive deceleration. In contrast, prediction-based longitudinal action initiates early braking to establish a safe gap, while prediction-based lateral action performs a lane change when a safe adjacent lane is available. Both predictive strategies avoid collisions in the evaluated records. Compared with reactive driving, peak deceleration before impact is reduced by 31.5% with longitudinal action and 91.0% with lateral action. Additionally, an ISO 2631-1-based horizontal vibration estimate is used to provide a ride comfort assessment. Overall, the results highlight the potential of lane-change prediction to support safer and smoother driving through proactive decision-making, with additional insights into ride comfort provided by vibration-based estimates.
1 INTRODUCTION
Unexpected merges can leave little room to brake or avoid, while predicting them gives an autonomous vehicle time to choose a response. This study compares reactive driving with prediction-based braking and lane avoidance, considering both safety and ride comfort.
1 INTRODUCTION
- Prediction gives the ego vehicle time to brake for a safe gap or change lanes when an adjacent lane is available.Figure 1 depicts a target vehicle merging into the ego vehicle’s lane, with red for collision, blue for braking, green for lateral avoidance, and transparent vehicles for future positions.Source: §1 · p. 1, Figure 1 · p. 1
- Braking and lateral avoidance have different motion profiles, so responses should be evaluated for passenger comfort as well as collision avoidance.Discomfort depends on motion magnitude, direction, and timing.Source: §1 · p. 1
- The study compares reactive driving, prediction-based longitudinal control, and prediction-based lateral control in CARLA using response timing, collisions, acceleration, jerk, and an ISO 2631-1-based horizontal vibration estimate.Source: §1 · p. 2
2 RELATED WORK
Prior work covers lane-change prediction, decision-making under uncertain intentions, and measures of passenger discomfort. The safety and comfort trade-offs between anticipatory braking and lateral avoidance in response to predicted lane changes remain insufficiently explored.
2 RELATED WORK
- The safety and comfort trade-offs between anticipatory braking and lateral avoidance in response to predicted lane changes remain insufficiently explored.The study addresses this gap by comparing both prediction-based strategies with reactive driving in CARLA across collision outcomes, response timing, vehicle dynamics, and ISO-based ride comfort.Source: §2 · p. 2
3 METHODOLOGY
The methodology combines a knowledge-graph and Bayesian lane-change predictor with risk-based lane selection and vehicle controllers. CARLA simulations compare three driving configurations, while ride comfort is estimated from frequency-weighted horizontal acceleration over a common pre-collision interval.
3.1 Lane-Change Prediction
- The lane-change predictor converts 12 vehicle and traffic features into linguistic categories and knowledge-graph triples, learns TransE embeddings from highD and CRASH data, then uses Bayesian inference to predict left lane change, lane keeping, or right lane change.Figure 2 schematically shows the sequence from measurement categorization to graph embeddings and Bayesian maneuver prediction.Source: §3.1 · p. 2, Figure 2 · p. 2
3.2 Decision Making
- The decision system assigns a predicted target vehicle to its destination lane before the maneuver, scores lanes by positive finite closing TTCs, and selects the lowest-risk available lane.IDM–PID controls longitudinal motion; MPC tracks a fifth-order lane-change trajectory, while reactive driving considers the target only after it enters the ego lane.Source: §3.2 · p. 2
3.3 Experimental Setup
- The CARLA evaluation uses three runs per configuration, with a two-lane road for reactive driving and anticipatory braking and a three-lane road for lateral avoidance.Figures and numerical results come from a representative run; vehicle dynamics and comfort are compared over the three seconds before the reactive collision.Source: §3.3 · p. 3
3.4 Ride Comfort Assessment
- Ride comfort is estimated by frequency-weighting longitudinal and lateral accelerations with ISO 2631-1 Wd and combining them as horizontal vibration.The estimate uses simulated vehicle acceleration as a proxy for passenger-seat motion and omits full representation of vertical, rotational, and high-frequency vibration, so it is indicative rather than a complete ISO assessment.Source: §3.4 · p. 3
4 RESULTS AND DISCUSSION
In the evaluated scenarios, predictive responses avoid collisions and reduce abrupt vehicle motion. Lateral avoidance also has the lowest ISO-weighted horizontal acceleration estimate, though that comfort assessment is indicative.
4.1 Safety and Response Timing
- The reactive case collides at a minimum TTC of 0.015 s, while neither predictive configuration collides; reactive braking starts 0.97 s before impact versus about 3.5 s for predictive responses.The selected scenario anticipates the lane change about 2.5 s before lane-marking crossing; a low TTC during lateral motion does not necessarily signal imminent collision because TTC omits lateral separation.Source: §4.1 · p. 3
4.2 Vehicle Motion and Ride Comfort
- Reactive collision peaks at 18.8 m/s2 deceleration; prediction cuts pre-impact peak deceleration 31.5% with braking and 91.0% with avoidance, while avoidance yields the lowest combined RMS acceleration and jerk.Lateral avoidance nevertheless raises RMS lateral acceleration from 0.037 to 0.388 m/s2 and lateral jerk from 0.095 to 0.372 m/s3 versus prediction-based braking.Source: §4.2 · pp. 3–4, Table 1 · p. 4, Figure 9 · p. 5
4.3 ISO-Based Comfort
- ISO-weighted combined RMS acceleration is 1.52 m/s2 reactive, 1.42 with anticipatory braking, and 0.53 with lateral avoidance, reductions of 65% and 63% versus reactive driving and braking.The estimate uses horizontal vehicle accelerations rather than passenger-seat vibration, so the ISO discomfort ranges are indicative rather than definitive.Source: Table 2 · p. 5, §4.3 · p. 5
5 CONCLUSION
Lane-change prediction supported earlier braking or lateral avoidance, improving the evaluated safety and comfort outcomes while requiring broader validation.
5 CONCLUSION
- Peak longitudinal deceleration fell by 31.5% with anticipatory braking and 91.0% with lateral avoidance; combined RMS jerk fell by 58.7% and 87.8%, respectively.Compared with braking alone, lateral avoidance further reduced peak deceleration by 86.9% and combined RMS jerk by 70.5%.Source: §5 · p. 6
- Prediction enabled responses approximately 2.5 s earlier than reactive driving, preventing the collision observed in the reactive scenario.The results support selecting early braking or lateral avoidance according to surrounding traffic conditions and available space.Source: §5 · p. 6
- Lateral avoidance had the lowest frequency-weighted acceleration and most favorable estimated comfort level despite introducing additional lateral motion.Source: §5 · p. 6
- Findings are limited to simulated scenarios and vehicle-level acceleration measurements; diverse traffic tests and passenger-seat vibration measurements are needed for validation.Source: §5 · p. 6